Assessment 101
Unlike many tech firms, which keep the brains of their operations sequestered in dimly lit cubicle farms veneered with the orange dust of long-gone Cheetos stuck to spilled Mountain Dew, 3TIER lets its scientists out in public. Many of the questions Scott Eichelberger, our assessment tsar, fields when he’s staffing the 3TIER booth at meetings have to do with how we perform our weather modeling magic, and where the data comes from. In fact, I had the same questions. Lucky for me, Scott is a kind and patient person, and a gifted explainer.
The most surprising thing I learned from Scott is that the computer model 3TIER uses for assessment, the Weather Research and Forecasting (WRF) model, is a free program produced and constantly improved by government and research institutions. I could download a copy for my laptop. The data set with which we begin each model run also is freely available, a gift to the planet’s people from the National Centers for Environmental Prediction and the National Center for Atmospheric Research (NCEP-NCAR).
Of course, amateur that I am, the output from my Home Wind Assessment Kit would be useless hokum, indicating that glaciers had just retreated from my yard.
Mad Skills
“There’s lots of skill in running WRF in the correct configuration so you get a good result,” Scott says. “You need lots of experience working with WRF and meteorological knowledge to set up a good model run.” You also need to remember to input important stuff, like data that represents mountains and lakes and other features that influence the atmosphere.
Even if I did have the experience to run WRF at home, the computational demands would melt my MacBook. 3TIER has as much computing power as a respectable university atmospheric sciences department, and pretty much continually adds more CPUs to its cluster. “We used to by five or six computers at a time,” says Jim McCaa, who keeps it all humming. “Now we’re buying 80 at a time.”
NCEP-NCAR’s data is for atmospheric conditions at six hour increments around the whole planet, going back about 50 years. It is constantly refreshed as observational data comes in from various airport towers, weather balloons, satellites, and radar stations. NCEP-NCAR puts all these mathematical values into a computer model that checks for weird numbers that might result from measurement errors and “smoothes” things that may vary widely from time point to time point, like wind speed and temperature. The output is a data set that represents atmospheric conditions on the globe as if it were covered by a grid of boxes 2.5 degrees longitude tall by 2.5 degrees latitude wide. Roughly, at middle latitudes that’s a resolution of about 277 kilometers (172 miles).
Which is useless to a guy in Nebraska looking to plant a wind farm. He wants to know what the wind is like every couple hundred feet at a specific latitude and longitude.
Heavy Math
But the NCEP-NCAR data is invaluable to Scott and the assessment staff because it provides a starting point, or what they call the boundary conditions, for the WRF model. The WRF model is one of several so-called numerical weather prediction models, computer programs capable of solving a bevy of equations that represent how a parcel of air behaves in the atmosphere, and how parcels interact with each other, depending on the conditions of the atmosphere: temperature, pressure, humidity, and wind speed. It can also account for how features on the ground will influence each parcel, and for solar radiation and the rotation of the globe. The series of math problems that represent these conditions would be tough to solve on paper, Scott points out, because each equation is linked to the equations that represent the conditions in the parcels above, below, and beside a given parcel. It may take several weeks of high-speed computing, but 3TIER’s supercomputers are up to the task.
Several model runs are required to refine the data to a resolution that’s useful to a Nebraska wind farmer. Each run’s output provides the boundary conditions for the next model run. The result is called a nested grid: a grid with a smaller grid in one box, and a smaller grid in one of those boxes, and a smaller grid in one of those boxes … . With each new nest, the data points on the grid are closer together by a factor of three.
Hi-Rez
Different clients ask for assessments of different resolutions, over areas large and small, going back 10, 25, even 40 years. As of this post, we can produce an assessment with a horizontal resolution as fine as 500 meters, but Scott’s team is working to create even higher resolution model output.
Computer models are not infallible, and one known issue with this method is that the model doesn’t completely capture the true complexity of the real atmosphere. To bring the model’s output in line with real conditions, we use observational data that clients collect from towers at their site and a method called model output statistics (MOS) correction. MOS adjusts the raw WRF model output to better match the observational, on-site data. At last, the final, MOS-corrected data set is delivered to the client. Some firms do it cheaper, but Scott and the assessment team prefer to do it right.
Tags: Assessment, WRF
Author: Leah Kauffman
